Briefing note to band council – back today with feedback Housekeeping
MIDTERM – FEB 17!
•All readings* covered, with emphasis on portions I presented in class
•Format: short & medium answers, a long answer
Peterman 2004 ICES Journal of Marine Science, 61: 1331-1343
This is process and observation error, which combine to ``Mortality Limit Uncertainty`` in today`s grizzly bear example
Uncertainty in Management
1. Process Error: natural variation
2. Observation (measurement) Error:
Most error refers to uncertainty, not mistakes
Important implications because the consequences of ignoring uncertainty can be great
3. Model Selection Error:
use of “more wrong” model
Why Grizzly Bears?
• Species with vulnerable life history
• High Conservation Value
• Most human-caused mortality is trophy hunting via
special permits (so, “manageable” by Ministry of Environment)
• Contentious debate, especially in coastal BC, and without sufficient facts
Confronting error and uncertainty in wildlife management: grizzly bear mortality in British Columbia, Canada
•
Much unknown/uncertain about populations
Our objectives
• Examined 2 subsets of uncertainty
– Implementation error
– Process & Observation error – “mortality limit uncertainty”
• First, distinguish between mortality limits and mortality targets
Limit vs. target
• Mortality Limit:
– from biology of species
– has Process Error about it; nature is variable
– our estimate adds Observation error; hard to measure basic population parameters well
– exceeding it should be avoided if you do not want population declines
• Mortality Target: -
The two are often used interchangeably in wildlife management, but are fundamentally different
Target setting by Ministry….though they are really estimated limits (they did not differentiate)
Total mortality target/limit = (Population Estimate × Annual Allowable Mortality)
Targets set for each Grizzly Bear Population Unit (GBPU) for each allocation period
• 2001-2004 • 2004-2006 • 2007-2011
− Estimated Unreported Mort.
Our “audit” (to assess implementation error)
•
Difference between allowable (target/limit) vs.
reported (actual) human-caused mortality
Combined (female + total)
overmortality events
across about 50 GBPUs, 2001-2011
2001-2003 2004-2006 2007-2011
• Actual mortality follows a distribution around the target (we now know about implementation error in action)
Target
Simulations: Confronting Uncertainty
•
Part 1. How does process/observation
uncertainty in population parameters propagate
to uncertainty in mortality limits?
Total mortality target/limit = (Population Estimate × Annual Allowable Mortality) − Estimated Unreported Mort.
Varied Parameter 1: Population Estimates
Ministry’s own estimates have changed -89% to +130% in less than 10 years
We varied population estimates by up to ± 40% in both
directions from values the Ministry uses
Varied Parameter 2: Annual Allowable Mortality (i.e.,
what proportion can we kill to ensure the population will not decline?)
Maximum:
•
5.7% (Bunnel & Tait)
•
2-3% (Sidorowicz & Gilbert 1981)
•
4.9% (McLouglin 2003) (0% in “low quality”
habitat)
•
7.8% (McCullough 1981 and 1986)
Ministry currently uses a max is 6%; we varied from of 4% to 8%
Varied Parameter 3: Unreported Mortality
•
Could be as high as 25% of human-caused
mortality (Peek
et al.
1987)
•
In Alaska, only half of grizzly hunting was
reported (Miller 1990)
•
26% of kills from McLellan et al. 1999
meta-analysis of BC populations
•
We
varied
unreported mortality by
up to ±
50% in both directions
from values Ministry
uses
Simulations (to confront Process and Observation Error..i.e., uncertainty)
•
Simulate limits by
randomly varying
each
parameters
within ranges of uncertainty
:
– Population Estimate: ± 40% around what Min uses – AAM: ± 2% around what Min uses
– Unreported Mortality: ± 50% around what Min uses
Repeat 1000 times to create distribution of LIMITS for each GBPU (i.e. run those equations 1000 times (ie, ‘runs’) & get 1000 different answers for the limit)
Total mortality target/limit = (Population Estimate × Annual Allowable Mortality) − Estimated Unreported Mort.
Distribution of simulated LIMITS from 1000 runs Previous Limit (Target) used by Ministry Distribution of actual limit Limit (# of bears) % oc curr en ce in simula ti on runs
Part 2. Simulations to confront Implementation Error
•
Implementation error similarly follows a
distribution
•
Constructed from historic relationship
between allowable and reported mortality
Randomly sample these values 1000 times to create distribution of TARGETS for each GBPU
Target
Distributions of actual mortality based on TARGETS from 1000 runs Distribution of actual mortality around target (implementation error) % oc curr en ce in simula ti on runs Target (#s of bears)
“Probabilistic” framework for setting NEW, SAFER targets that incorporate all sources of estimated error and uncertainty
• Part 3 is overlaying Parts 1 & 2 together
• Incorporates both implementation error (around targets) and
Probabilistic framework for setting
targets
•
For a given %
probability of
overmortality (x),
set target such
that
x % of limit
distribution
overlaps with
target
distribution
•
Can shift target
based on
acceptable risk
Target Limit Probability of overmortality Previous Limit (Target) used by Ministry Distribution of actual limit New Target Distribution of actual mortality around target (implementation error) Target/Limit (#s of bears) % oc curr en ce in simula ti on runs5% probability of overmortality
New Target Previous Limit (Target) used by Ministry 5% Distribution of actual limit Distribution of actual mortality around target (implementation error) % oc curr en ce in simula ti on runs Target/Limit (#s of bears)10% probability of overmortality
10%
We can move the new ‘target range’ to find a level of harvest ‘we’ like;
New Target Previous Limit (Target) used by Ministry Distribution of actual limit Target/Limit (#s of bears) Distribution of actual mortality around target (implementation error) Distribution of actual mortality around target (implementation error) % oc curr en ce in simula ti on runs
5% probability of overmortality
New Target Previous Limit (Target) used by Ministry 5% Distribution of actual limitDifference in target required to bring the probability of overmortality (i.e., target > limit) to below 5%
Distribution of actual mortality around target (implementation error) % oc curr en ce in simula ti on runs Target/Limit (#s of bears)
Using this new approach - How would changes to hunting affect the probability of overmortlity (under 5%)?
-reducing hunting by half would have reduced overmortality events by 86%
-completely eliminating
hunting would have reduced overmortality by 96%
Confronting uncertainty is both necessary and….
useful in that it derived a better scientific way to set “mortality targets” Yay, science…we now know how many bears “should” be killed
for trophy!
Resource management, however, is also about incorporating values
Bear hunting ban declared by 10 B.C. First Nations
But provincial government says only it has the authority to issue such a ban The Canadian Press Sep 13, 2012 6:46 AM PT